soil moisture conservation in very dry and sandy conditions

[har917]

Following analysis in the coupled model we have an identified a source of moisture imbalance in the default soil_snow scheme.

At the moment

• cbl_Surface_wetness (line 66) (and cbl_init_wetfac) prevent evaporation from happening if %wb < %swilt/2
• surbv (line 54) prevents %wb falling below 0.01

This combination allows evaporation from dry soils (when '0.02>%wb>%swilt/2and%swilt < 0.02) which then leads tosurfbvcreating water to maintain%wb` at the hard-wired lower limit. To prevent this we need link the two limits together.

My suggestion

• create a new constant wilt_limitfactor = 2.0 in cable_phys_constants
• in cbl_Surface_Wetness and cbl_init_wetfac change the relevant lines from soil%swilt/2.0 to soil%swilt/wilt_limitfactor
• in surfbv change ssnow%wb(:,k) = MAX( 0.01_r_2, MIN( ssnow%wb(:,k), xxx ) ) to ssnow%wb(:,k) = MAX( soil%swilt/wilt_limitfactor/2.0, MIN( ssnow%wb(:,k), xxx ) )

This then allows a single point for modification of the relevant lines.

This change will only impact soils where %swilt < 0.02 which, effectively, means almost pure sand. Until recently soils with such low values %swilt were not encountered in our standard configurations but the latest UM soil ancillaries do comprise regions where this occurs.

---

We may also need to consider what happens with alternate formulations for %wetfac which can allow evaporation at very low values of %wb

[ccarouge]

I would define wilt_limitfactor in cable_other_constants which already contains other "arbitrary" limits. Ideally, I would rather define wilt_limitfactor = soil%swilt/2.0 but that would mean having a run-specific and location-specific "constant" and I'm guessing that wouldn't play well with JULES. That would mean defining it somewhere else, and I'm not certain about adding it to soil%, as a derived parameter.

Anecdotally, would the name wilt_limit be specific enough but shorter to type?

[har917]

While the latest fix to the AM3 code see #127 should avoid the occurrence of this %swilt related water imbalance - the solution isn't robust in that the fundamentals still exist (we've just moved the problem to a part of parameter space that we're unlikely to encounter with soil_snow and the updated UM soil ancillaries). It could well be encountered with other combinations of parameterisation schemes (such as Mark Decker's implementation of the alpha-form for soil evaporation)

A better solution would be/may be to shuffle water around the soil column if the model determines it needs to - i.e. diagnose how much water would be needed to keep the surface soil layer at a certain level of water content and extract that amount from the deepest soil layer if needed.

In addition - for global conservation in the coupled model - we should probably be capture any water that we do need to add as part of the %wb_lake term so it can be used in the river outflow scaling code.